Quality of Education and Public Resource Allocation in Brazil
Ricardo S. Freguglia1, Mônica A. Haddad2, and Cláudia Gomes3
Abstract: The main objective of this paper is to examine whether Brazil’s scarce public resources were being allocated
in a coordinated manner during the period 2003-2009. The overall research question it attempts to answer is as
follows: what are the relationships between quality of education, educational spending, and BF cash transfer? Prova
Brasil, a national test administered by the Brazilian Ministry of Education, is used as a proxy for measuring quality of
education. Higher average scores in Prova Brasil indicate better quality of education. Fundef data is used to measure
public spending on education. A fixed effects approach is used, including variables at the school and municipal levels.
Results indicated that the hypothesis about higher BF allocation coinciding with lower Prova Brasil average scores is
validated just to the early elementary math education (Mathematics 4th grade) and to the Portuguese 8th grade
education. This may be due to the fact that municipalities’ skills to implement BF vary vastly, indicating that there is
not a consistent municipal behavior with regards to BF. When considering Fundef resources, the hypothesis about
higher Prova Brasil scores should be receiving higher government spending on elementary education was supported. In
both situations, the magnitudes of the coefficients were very small, indicating that the returns to education that are
expected from public spending are not yet very relevant for the Brazilian case. In general, these results may be related
to difference in municipal performances in administering public spending.
KEYWORDS: Conditional cash transfer, basic education, public investment, fixed effects, human
capital, Brazil.
JEL Code: I25, I28
Resumo: O objetivo central desse artigo é examinar se os escassos recursos públicos estão sendo alocados de maneira
coordenada no Brasil no período de 2003 a 2009. A pergunta principal que o artigo busca responder é: qual a relação
entre qualidade da educação, gastos educacionais e transferência de recursos do Bolsa Família? A proficiência dos
alunos na Prova Brasil, do Ministério da Educação, é usada como Proxy para medir a qualidade da educação. Notas
mais elevadas na Prova Brasil indicam melhor qualidade da educação. Dados do Fundef são usados para medir o gasto
público em educação. A abordagem de efeitos fixos é adotada, incluindo variáveis ao nível da escola e dos municípios.
Os resultados indicam que a hipótese de que maiores gastos no Bolsa Família coincide com menores notas médias no
exame da Prova Brasil é validada apenas para a primeira etapa da fundamental em matemática (4a ano) e para o 8o ano
em Português. Isso pode ser devido a grande variabilidade na habilidade dos municípios em implementarem o BF,
indicando que não existe uma conduta municipal padronizada relativa ao BF. Ao considerar os recursos do Fundef, a
hipótese de que notas mais elevadas da Prova Brasil deveriam estar recebendo maiores gastos governamentais sobre a
educação básica é sustentada. Em ambos os casos, as magnitudes dos coeficientes estimados é muito pequena,
indicando que os retornos a educação que são esperados em decorrência dos gastos governamentais não são muito
relevantes para o caso brasileiro. Em linhas gerais, sugere-se que tais resultados podem estar associados as diferenças
de desempenho dos municípios na gestão dos recursos públicos.
PALAVRAS-CHAVE: Bolsa Família, Educação Fundamental, Investimentos Públicos, Efeitos Fixos,
Capital Humano, Brasil.
Área ANPEC: Economia Social e Demografia Econômica
1
Assistant Professor, Federal University of Juiz de Fora, Department of Economics.
Associate Professor, Iowa State University, Department of Community and Regional Planning.
3 Master in Economics, Federal University of Juiz de Fora, Department of Economics .
2
1|Page
1. INTRODUCTION
Brazil has been in the headlines of international media because of its high economic growth rate and
its investments in social development. In the context of the global economy, despite the current economic
crises, Brazil is growing at a faster rate than most other countries. This growth is attracting many investors
and new businesses to its territory. Additionally, under President Lula’s leadership (2002-2010), social
investments were a priority, allowing many Brazilians who were living in poverty to have a better quality of
life. According to the World Bank (2012), the poverty rate in Brazil rose to 36 percent in 2003, but has
sharply declined since then to 27 percent in 2006 and 21 percent in 2009. The population earning less than
Purchasing Power Parity of $1.25 per day followed a similar pattern, rising to 11.2 percent in 2003 and
falling to 6 percent in 2008, before a very slight rise to 6.1 percent in 2009.
Most of the social investments initiated during President Lula’s tenure had the ultimate goal of
forming human capital. These investments happened through a conditional cash transfer program named
Bolsa Família (BF). Under President Dilma Roussef’s administration (2011-2014), the Brazilian government
continues to allocate public resources for the BF program. The program is based on direct cash transfer to
poor families who agree to keep their children in school. The program also provides recipient families with
basic health care. According to Ministério do Desenvolvimento Social (2012) BF benefited 8.7 million
families in 2005, at a cost of R$5.7 billion; more than 21 percent of the federal budget was allocated for all
social programs that year. In 2010, approximately 12.8 million families benefited at a cost of R$1.2 billion.
Because of BF requirements, this increasing number of beneficiaries leads to increasing enrollment in public
schools. Within this context, quality of education is an important part of the Brazilian economic growth
process. In other words, children who receive a high quality education will be more likely to join the
qualified work force that a growing country like Brazil is in need of.
When comparing economic growth, quality of education, and educational spending in a few Latin
American countries, some interesting findings arise. Table 1 displays this comparison from 2006 to 2009, in
which economic growth is described using GDP annual percentage growth, quality of education is described
using the UNESCO index EDI4 (‘education for all’ EFA development index), and educational spending is
described using the annual percentage of GDP that is allocated to education. Brazil was the only country that
did not have an increasing trend in EDI from 2006 to 2008. In addition, Brazil had the lowest EDIs when
compared to its three neighbors. When focusing on GDP, the effect of the 2008 world economic crises is
visible: all countries had a decrease in their GDPs from 2008-2009. In fact, the GDPs of Brazil and Chile
decreased between 2008 and 2009. Concerning educational spending, Brazil had the highest GDP percent
allocated to education in 2006 and 2007, and spending continued to increase in 2008 and 2009.
From the economic growth perspective, the idea that growth goes hand in hand with investments in
human capital in widely accepted. For instance, Baldacci et al (2008) demonstrated that “both education and
health spending have a positive and significant impact on education and health capital, and thus support
higher growth” (p.1317). Table 1 shows that this was happening in Brazil from 2006 to 2008; i.e. public
spending on education steadily increased. However, given this, one might expect Brazil to have higher EDI
values and an increasing trend in EDI. This discrepancy may be due to the need for a time lag for the index
to capture the changes that the investment should promote. This may also be due to the high enrollment rates
that are taking place in public schools because of BF.
4
The United Nations Educational, Scientific and Cultural Organization (UNESCO) index is a composite index, which includes four variables in
its calculation: universal primary education, measured by the primary adjusted net enrolment ratio; adult literacy, measured by the literacy rate for
those aged 15 and above; gender parity and equality, measured by the gender-specific EFA index, an average of the gender parity indexes of the
primary and secondary gross enrollment ratios and the adult literacy rate; and quality of education, measured by the survival rate to grade 5. No
EFA was available for 2009.
2|Page
Table 1: Comparing Economic Growth, Quality of Education, and Educational Spending, 2006-2009.
Year
Chile
Argentina
Colombia
Brazil
EFA
Development
Index (EDI)
Annual %
change in GDP
Public spending
on education in
% of GDP
2006
--
0.956
0.905
0.901
2007
0.966
0.971
0.920
0.883
2008
0.968
0.972
0.929
0.887
Change
↑
↑
↑
↓
2006
5.69
8.47
6.70
3.96
2007
5.16
8.65
6.90
6.09
2008
3.29
6.76
3.55
5.17
2009
-1.04
0.85
1.65
-0.33
Change
↓
↓
↓
↓
2006
3.2
4.5
3.9
5.0
2007
3.4
4.9
4.1
5.1
2008
4.0
5.4
3.9
5.4
2009
4.5
6.0
4.7
5.7
Change
↑
↑
↑
↑
Source: UNESCO and World Bank
As noted above, two increasing trends of public investments are taking place in Brazil, in social
development and in education. Coordination in the allocation of these public resources would contribute to
the formation of human capital. Table 2 depicts specific total annual amounts of public resources that were
allocated to education and social development between 2003 and 2007, corresponding to the period of the
data below. The former is illustrated by data on Fundo de Desenvolvimento do Ensino Fundamental e da
Valorização do Magistério (Fundef), described in more detail below, and the latter is illustrated by data on
BF. The selected years correspond to the period of study of this paper. All percent changes were positive.
BF change from 2003 to 2005 was extremely large (790%) because this period captures the switch in the
federal budget allocation towards social development. These numbers indicate that public spending in Brazil
shifted towards social development; a very wise strategy for a growing economy known for its high social
inequality. These figures further emphasize the need to better understand the relationship between quality of
education, and public resource allocation to education and social development in Brazil, to assure that the
allocation of public resources is indeed contributing to the growth process.
Table 2: Public Resources Allocation – Fundef and BF.
Year
FUNDEF (in millions
Reais)
2003
12,815
---
2005
17,756
39%
6,853
790%
2006/2007
19,972
12%
8,625
26%
% Change
Bolsa Familia (in
millions Reais)
769
% Change
---
Note: FUNDEF data was available until2006.
Source: Tesouro Nacional and Ministério do Desenvolvimento Social.
The main objective of this paper is to examine whether Brazil’s scarce public resources were being
allocated in a coordinated manner during the period 2003-2009. The overall research question it attempts to
3|Page
answer is as follows: what are the relationships between quality of education, educational spending, and BF
cash transfer? Prova Brasil, a national test administered by the Brazilian Ministry of Education, is used as a
proxy for measuring quality of education. Higher average scores in Prova Brasil indicate better quality of
education. Fundef data is used to measure public spending on education. The research question was
answered using regressions that treated Prova Brasil scores as the dependent variable and BF and Fundef
spending (along with numerous controls) as the independent variables. The results of the regression also
allowed us to address the paper’s two research hypotheses: 1) municipalities with higher level of BF
allocation also had lower scores in Prova Brasil; and 2) municipalities characterized by higher Prova Brasil
scores also received higher government spending on elementary education. The first hypothesis is based on
the fact that BF is increasing enrollment in public schools, but the public school system may not be able to
provide resources needed to meet the needs of more students. The second hypothesis is based on the idea
that higher government spending on education should result in better education performance.
The paper is organized as follows. Section 2 presents a literature review, and section 3 describes the
variables used in this paper and the methodology. Section 4 presents the estimation results. The final section
contains concluding remarks, including policy recommendations, limitations of the study, and ideas for
future research.
2. LITERATURE REVIEW
This review focuses on three main topics. First, BF started being implemented in Brazil in the early
2000’s and there is extensive published literature about this program, exploring a variety of topics. Second,
given that Brazil is investing a large portion of its GDP in education, that the ultimate goal of BF is to form
human capital, and that Brazilian economic growth is attracting many investors and new businesses, it is
important to understand the relationship between education and economic growth. Third, examining work
related to educational spending in Brazil contributes to a better understanding of our regressions results.
a) Bolsa Família
The first step for a municipality to receive BF funding is to establish an agreement with the federal
government. The municipality then becomes responsible for registering families interested in becoming BF
beneficiaries. This registration process happens through an electronic system, named Cadastro Único, which
provides the federal, state, and municipal governments with access to the information. Funding allocation to
BF beneficiaries varies according to the family poverty level, the number of children in the family, or the
presence of a pregnant or nursing woman in the family. Once the agreement is established, each municipality
is responsible for managing and distributing BF funds in its territory. Fried (2012) showed “strong evidence
that BF is distributed in a programmatic manner” (p. 1049) indicating that its distribution process is not
dependent on political criteria such as clientelism and political interference in how transfers are distributed.
Overall, most of the studies indicate that the program is allowing many Brazilians who were living in
poverty to have a better quality of life. Studies that analyzed the relationship between BF and equality show
that poverty and inequality were decreasing in Brazilian municipalities because of BF (Soares et al., 2006;
Tavares et al., 2009; Landin, 2009; Soares et al., 2010; Vale et al., 2010; Machado et al., 2011). Other
studies that focus on BF and educational issues indicate increased enrollment and attendance, and decrease in
dropouts (Bourguignon et al., 2003; Haddad, 2008; Kassouf and Glewwe, 2008; Romero and Hermeto, 2009;
Vale et al, 2010, Machado et al, 2011). Specifically focusing on student proficiency, two studies found a
negative relationship as a result of increased BF spending (Liso, 2010; Camargo, 2011). Moreover, studies
on child labor have found decreasing trends (Ferro and Kassouf, 2005; Pedrozo, 2007), but also no change,
and limited effects (Cardoso and Souza, 2004; Machado et al, 2011). Finally, research on health issues had
4|Page
mixed findings (Paes-Sousa et al., 2011; Andrade et al., 2012). Several of these studies are described further
below.
Concerning education, Haddad (2008) showed the importance of BF in increasing enrollment in
public schools, and confirmed that public resources allocated to BF were contributing to greater social
equality. Similarly, Glewwe and Kassouf (2008) estimated that the program will raise school enrollment
among children in eligible families by 13 percent, will lower dropout rates by 1 percent, and will raise grade
promotion rates by 2.3 percent in the long run. With regards to student performance, Liso (2010) showed a
negative relationship between 8th grader proficiency and the program, pointing out that diminished quality of
education in public schools, related to increasing enrollment, was a consequence of BF. In the same
direction, Camargo’s (2011) results showed a negative relationship between proficiency and BF. But the
author indicated that the schools included in his study already had lower performance, prior to BF’s creation.
In addition to increasing school enrollment, ideally BF should also diminish child labor and income
inequality. Indeed, results of a study by Pedrozo (2007) indicated that including families in conditional cash
transfer programs led to the reduction of such labor. Conversely, in a study of PNAD (Pesquisa Nacional
por Mostra de Domicílio) data from 2004-2006, Machado et al.(2011) found that BF had only limited effects
on curbing child labor, but did slightly boost school attendance. However, they did see improvements in
income inequality, showing that beneficiaries were less likely to work in the informal economy and earned
more by the end of the two-year period studied.
With regards to BF health service provision, on one hand, Paes-Sousa et al. (2011) found that
children in families that benefit from BF were 26 percent more likely to be of normal height for their age,
indicating that they have greater access to adequate food. On the other hand, even though BF beneficiaries
have access to child immunization services, Andrade et al. (2012) found that the program had little effect on
child immunization rates among beneficiary families, despite a higher rate of immunization among children
under 6 months of age.
In summary, despite the fact that BF has been the topic of various studies, there has been no
systematic study that yet relates the coordination of public resources allocation in social development and
education, and quality of education.
b) Education and Economic Growth
In many countries educational spending has proven to be effective means of accumulating human
capital. Likewise, various approaches have demonstrated how education may affect economic growth. First,
some scholars (see Lucas, 1988; Mankiw, Romer, and Weil, 1992; Texeira and Fortura, 2004; Oketch, 2006;
Fleisher, 2011) empirically showed that education fostered human capital, which increased labor
productivity, and as a consequence, moved economic growth to a higher level. Second, various studies (see
Nelson and Phelps, 1966; Lucas, 1988; Aghion and Howitt, 1998; Ranis and Ramirez, 2000; Wolff, 2000;
Lin 2003; Benhabib and Spiegel, 2005; Park, 2008) indicated that education boosted both innovation
capacity and new knowledge about technology, products, and processes leading to growth. Third, the fact
that education can facilitate spillovers of knowledge and promote skill-based technical changes (Fleisher,
2011) is well documented (see Acemoglu 1996 and 1998; Ciccone and Peri, 2006).
Investments in education are related to differences in levels of development between countries and
regions. These global differences are illustrated by the work of Hanushek and Wossmann (2007). By
comparing education in different countries, they found that educational deficits were larger in developing
countries, where programs focused only on school enrollment and attendance, and not on quality of
education. They showed that in order to decrease the economic deficit between countries, changes in the
structure of educational institutions in developing countries should take place. Barro (2001) also contributed
to this discussion, indicating that program measures that focused on quality of education (evaluated by tests
of cognitive abilities) were more important for economic growth than measures such as enrollment and
5|Page
attendance. Along these lines, Baldacci et al (2008) found that public spending on health and education was
positively and significantly related to human capital (measured using the sum of the gross primary and
secondary enrollment rates), resulting in economic growth.
There are, however, some studies that contradict the ones presented above. Teles and Andrade (2008)
demonstrated that there was no consensus in empirical studies that focused on the relationship between
government spending on education and economic growth. Additionally, Blankenau and Simpson(2004)
showed that even when public expenditure promotes human capital, economic growth may not increase. In
summary, the findings reviewed above stress the need to better understand how the process of human capital
formation is evolving in Brazil.
c) Educational Spending in Brazil
The system that finances education in Brazil is very complex; various funds work together, a large
combination of taxes are used to accumulate resources, and resource allocation is based on many criteria.
Therefore, for the purpose of this paper, a narrowed approach is presented focusing only on Fundef. Recall
that Fundef spending is the variable used in this paper to capture public investment in education. The 1988
Brazilian Constitution created Fundef, which lasted from 1997 to 2006.In Brazil, the K-12 system (educação
básica) is divided into two levels. The elementary school (ensino fundamental) corresponds to grades 1 to 9
and the high school (ensino médio) corresponds to grades 10 to 12. Fundef resources were targeted only to
elementary schools. Fundef was replaced by the Fundo de Manutenção e Desenvolvimento da Educação
Básica e de Valorização dos Profissionais da Educação (Fundeb). Fundeb was created in 2007 and is
expected to end in 2020, and its resources are being target to both, elementary school and high school.
Fundef allocation process was based on automatic transfers to municipalities and states, according to
enrollment in elementary school. When Fundef was active, 60 percent of all education resources in the
country had to be allocated to maintain and develop elementary school. These resources were gathered in
state funds, and then, allocated to municipalities and state according to enrollment in elementary school.
This fund had a combination of resources from municipalities and states, including a variety of taxes. In
addition, the federal government would also contribute to the fund, assuring that a minimum per student was
reached. In other words, the federal government was committed to complement the resources in order to
assure a minimum value per student. This system allowed poor municipalities to have a minimum to spend
on students. On the other hand, rich municipalities, in addition to Fundef, also had other resources to spend
on students.5
A few studies prior to this one also examined Fundef. Davies (2006) developed an overview of the
Brazilian education financial system and found that Fundef did not bring new resources to the system and did
not improve teachers’ salary. Instead, the fund only redistributed a portion of the taxes between the state
government and municipalities, allowing some municipalities to receive more funding and others to receive
less. According to Davies (2006), the tax inequality that exists between the different levels of governments
in Brazil inhibited the formation of a national system of education spending with good standards and quality.
Using Fundef data, Mello and Hoppe (2005) studied the evolution of public expenditure on education
in Brazil during 1991-2002. When comparing Brazil to Organization for Economic Co-operation and
Development (OECD) countries, they found that Brazil invested a higher share of its budget in public
education programs. Despite the high spending on education, Brazilian students performed poorly,
indicating a problem in the quality of investments. Indeed, these international comparisons showed that
6|Page
some countries with lower public spending, achieved better outcomes than Brazil. The authors pointed out
that these discrepancies could reflect a lack of efficiency (Mello and Hoppe, 2005).
Gordon and Vegas (2005) found that increased spending from Fundef was associated with smaller
class sizes and small gains in enrollment. Conversely, they found no evidence that it has improved school
performance for most students, except perhaps for low-achieving and non-white students. In a study using
1998 Fundef data, Menezes-Filho and Pazello (2007) found that, overall, increases in teacher salaries due to
Fundef did little to improve performance in public schools. However, these raises did attract better new
teachers to some schools, and students of these teachers did have improved proficiency.
In summary, there are mixed results with regards to public spending in education in Brazil, making
the overall research question – what are the relationships between quality of education, educational spending,
and BF cash transfer? – an important one to be examined. The two research hypotheses – 1) municipalities
with higher level of BF allocation also had lower scores in Prova Brasil; and 2) municipalities characterized
by higher Prova Brasil scores also received higher government spending on elementary education –
specifically target this need.
3. DATA AND MODELS
a) Data
The data used in this study came from various sources. Table 3 describes all the variables used in the
models. Some variables were collected at the school level and others at the municipal level. The dependent
variables – average Prova Brasil scores – were obtained from the Insitituto Nacional de Estudos e Pesquisas
Educacionais Anisio Texeira (INEP). Prova Brasil started in 2005 and is administered by the Ministry of
Education every two years. In 2005, the test was administered in schools located in urban areas that had a
minimum of 20 students per grade. In 2007 and 2009 it was administered in schools located in urban and
rural areas that had a minimum of 30 students per grade.
The objective of Prova Brasil is to assess student proficiency in math and Portuguese for grades 4 and
8. The two hypotheses posed above were tested in four regressions using average scores for the four tests as
dependent variables. It was expected that the four dependent variables would have the same relationships to
the independent variables of interest, regardless of the test age and subject matter differences.
There were two independent variables of interest: 1) the amount of cash allocated through BF (in
Brazilian currency) in a municipality divided by the number of poor people in that municipality (BF_poor);
and 2) the amount of public spending on elementary education in a municipality divided by the number of
students enrolled in elementary schools in that municipality (Fundef). The BF data was gathered through the
Ministry of Social Development, and the number of poor to create the BF ratio came from Instituto
Brasileiro de Geografia e Estatística (IBGE) 2000 Census. IBGE defines poor people as the proportion of
individuals with per capita income below R$75.50, corresponding to half of the minimum wage in August of
2000 (Atlas do Desenvolvimento Humano no Brasil). Data on education spending was obtained from the
Department of the Treasury.
A variety of control variables were also included in the models. These were classified as student,
school, teacher, or municipality related, and identified from literature that examines student proficiency
(Menezes-Filho, 2007; Barros and Mendonça, 1998; Hanushek, 2006; Kilkenny and Haddad, 2008).
Characteristics of students who took Prova Brasil were obtained from INEP. The School Census (Censo
Escolar) developed by the Ministry of Education was used to get data for the characteristics of schools and
teachers in each municipality. Municipal characteristics came from the following sources: GDP per capita
came from IBGE; and income and health indices were calculated by the Federation of Industries of the State
of Rio de Janeiro. The income index included variables of average wage, generation and supply of formal
employment; and the health index included number of pre-natal and infant deaths.
7|Page
Table 3: Description of all Variables
Years
Unit of
analysis
Prova Brasil Math average
score - 4th grade (Math_4)
2005,
2007, 2009
School
Prova Brasil Portuguese
average score - 4th grade
(Port_4)
2005,
2007, 2009
School
Prova Brasil Math average
score - 8th grade (Math_8)
2005,
2007, 2009
Prova Brasil Portuguese
average score - 8th grade
(Port_8)
2005,
2007, 2009
Variable
Variable
Years
Unit of analysis
Proportion of public schools in a
municipality that had a computer
lab (Sch_comp)
2003, 2005,
2007
Municipality
School
Proportion of public schools in a
municipality that had a library
(Sch_lib)
2003, 2005,
2007
Municipality
School
Total number of students divided by
number of classes in elementary
schools (Stud_class)
2003, 2005,
2007
Municipality
2003, 2005,
2007
Municipality
GDP per capita (GDP)
2003, 2005,
2007
Municipality
Municipality
Dependent variables
Independent variables of
interest
Amount invested in
BolsaFamília - in Brazilian
currency - divided by number
of poor people (BF_poor)
Fundef spending divided by
number of enrollment in
elementary education
(Fundef)
School-related control variables
Teacher-related control variables
2003,2005,
2007 (BF
amount) &
2000
(number of
poor)
Municipality
2003,
2005, 2006
Municipality
Student-related control
variables
Proportion of teachers working in
public schools with undergraduate
degree (Teach_edu)
Municipality-related control variables
Proportion of students who
took Prova Brasil who lived in
a place with internet access
(Stud_int)
2005,
2007, 2009
School
Health index (HEALTH_Index)
2000, 2005,
2007
Proportion of students who
took Prova Brasil and worked
(Stud_wor)
2005,
2007, 2009
School
Income index (INCOME_Index)
2000, 2005,
2007
Proportion of mothers -of
students who took Prova
Brasil - who attended 4th
grade or higher (Mother_edu)
2005,
2007, 2009
School
Municipality
The number of observations of the variables displayed on Table 3 varied within a source, and also
from source to source. For instance, the number of schools that participate in Prova Brasil Math_4 in 2005
was 26,232, and in 2009 the number was 27,850. 5,564 municipalities were BF participants in 2007, and
Fundefwas present in 5,286 municipalities in 2006.In addition to that, all six different sources had missing
values, varying from variable to variable, and also from year to year. After merging these different data,
every school included in the econometric analysis appears in every year with no missing values. In other
words, the used data composes a balanced panel. As a consequence, the number of observations decreased as
table 5 below indicates.
The correlation matrix of the variables is displayed on Table 4. We used the fourth grade on math to
the sample from 2005-2009 to illustrate the mean behavior in this period. In general, the other variables
present their expected signal. The interest variables – BF_poor and Fundef – present a positive correlation
with the math grades. However, this behavior is not consistent when considering other grades. For instance,
BF_poor presents a negative correlation with the proficiency of students of the fourth grade.6 These
descriptive results deserves some additional exploration and motivate our investigation on additional factors
6
The correlation matrix to Math_8, Port_4, and Port_8 are available with the authors by request.
8|Page
which may contribute to the causal explanation between quality of education and public resource allocation
in Brazil.
Table 4: Correlation matrix
Math_4 lnBF_poor lnFundef Stud_int Stud_wor Moth_edu Sch_comp
Math_4
1
lnBF_poor
0.279
1
lnFundef
0.1504
0.4018
1
Stud_int
0.5698
0.0698
0.0802
1
Stud_wor
-0.4458
-0.1294
-0.0578 -0.3019
1
Mother_edu
0.3651
-0.0809
-0.073
0.4499
-0.316
1
Sch_comp
0.3042
0.1295
0.1198
0.3789
-0.1939
0.1939
1
Sch_lib
0.0461
0.0737
0.0634 -0.0353
-0.0741
0.029
0.1408
Stud_class
-0.0904
-0.2008
-0.1291
0.0717
0.0269
0.0671
0.0564
Teach_edu
0.3256
0.0765
0.0359
0.3712
-0.2344
0.2563
0.397
GDP
0.3113
0.0108
0.1532
0.4898
-0.2004
0.2758
0.3096
Health_Index
0.4884
0.198
0.0615
0.5727
-0.3106
0.3257
0.3931
Income_Index
0.3862
0.0847
-0.0869
0.5729
-0.2813
0.3789
0.315
Source: INEP, Tesouro Nacional and Ministério do Desenvolvimento Social.
Sch_lib Stu_class Teach_edu
1
-0.0197
0.174
-0.0248
-0.0009
0.0039
1
0.0469
0.0458
-0.0313
0.0943
GDP
1
0.2698
1
0.4523 0.4803
0.3406 0.5385
H_index I_index
1
0.6441
1
The dependent variables and the independent variables of interest deserve some exploration. Figure 1
displaysMath_4 for 2009, BF_poor for 2007 and Fundef for 2006. A common scale based on sample
averages was created in order to compare these three variables: low values were defined as those less than 90
percent of the sample average, and high values were defined as those greater than 110 percent of the sample
average. The spatial distributions of Math_4 and BF_poor presented some patterns of clustering. On one
hand, there was a spatial concentration of higher values of BF_poor in the Northeast region, and of lower
scores in Math_4 in the same region. On the other hand, the Southeast region was characterized by lower
values of BF_poor and higher values ofMath_4. Given that the Northeast is the poorest region in Brazil, and
the Southeast is the richest region, the distribution of BF_poor values in these regions is not surprising.
Furthermore, the coincidence of high BF spending and low Prova Brasil scores may be due to increasing
enrollment in public schools because of BF and inadequate resources to meet the needs of more students. The
spatial distribution of Fundef displayed an apparently random distribution in 2006, with no visual indication
of clustering.
Figure 1: Spatial Distribution of Dependent and Independent Variables of Interest
9|Page
Source: INEP, Tesouro Nacional and Ministério do Desenvolvimento Social.
Descriptive statistics, estimated using schools, for the dependent and independent variables of interest
are presented in Table 47. The total number of schools for 4th grade regressions is 8,833, and the total
number of schools for the 8th grade regressions is 7,546.In general, the average score for 8th graders is higher
than 4th graders for both, math and Portuguese. There is a consistent increase on the average of public
resource allocation for social development and education. The maximum values for Fundef differ a lot from
4th grade to 8th grade. An outlier school, located in São Vicente – RN – municipality with very high average
education spending was included in the 8th grade regressions, but not in the 4th grade ones.
b) Model Specification
The main goal of the econometric approach presented below is to capture the relationship between
quality of education, government spending on education, and BF allocation in Brazil during the period 20032009. The dependent variables were average test scores on Prova Brasil, which are used as a proxy for
quality of education8. Hanushek and Wößmann (2007) recommend that quality of education be measured by
standardized test scores rather than years of education to account for vast differences in educational quality
between countries, or even individual schools. In a survey of several studies, they found that higher
performance on tests translated to increased earnings, especially in the developing world. They also suggest
that higher cognitive skills, as measured by standardized testing, are also linked to lower repetition rates,
meaning that higher test scores could lead to higher educational attainment. While test scores cannot be
solely attributed to school performance, as factors like family background and living conditions have an
impact, these scores have significant implications for the economic performance of the students, and of the
countries that provided their education.
Student proficiency has been used as a proxy for quality of education in various studies (Barro, 2001;
Kilkenny and Haddad, 2008; Parankader et al., 2008; Liso, 2010). In particular, Prova Brasil was used as a
7
The full descriptive statistics, including all the independent variables considered in the regression, are presented in the appendix.
8
Natural logarithms were applied to the dependent variables and the independent variables of interest to allow for the measurement of elasticity.
By doing that, it was possible to capture how a change of one percent in the independent variable would affect the dependent variable.
10 | P a g e
dependent variable by Parankader et al. (2008) and Liso (2010). Barro (2001) found that standardized test
scores were positively correlated with economic growth, especially in science. An increase in science scores
by one standard deviation would result in a 1 percent rise in annual economic growth, whereas a rise in
educational attainment by one standard deviation would yield only a 0.2 percent gain in annual growth. This
echoes the work of Hanushek and Wößmann (2007) which suggests that the quality of education is more
important than the quantity of education in determining a nation's economic growth.
A panel data approach for Brazilian schools was used for the years 2005, 2007 and 2009. A span of
two years was applied between the dependent and independent variables as an attempt to capture any time
lags required for financial resources to start working. Specifically, the dependent variable 2005 Prova Brasil
had as its independent variables BF_poor and Fundef for 2003. The 2007Prova Brasil had both independent
variables from 2005, and the 2009 Prova Brasil had BF_poor for 2007 and Fundef for 2006 (last year of this
fund’s existence). Based on literature examining student proficiency (Menezes-Filho, 2007; Barros e
Mendonça, 1998; Hanushek, 2006), a variety of control variables were also included in the equation, as given
above in Table 3.
The panel data approach also allows for inclusion of unobserved municipal and school characteristics
that could influence Prova Brasil average scores. Examples of these characteristics could be public policies
that motivate reading, community engagement in early childhood education, the presence of NGOs that have
a strong capacity to be engaged in educational debates, school management, and other social factors that may
influence school performance in some way. When estimating a pooled ordinary least squares (POLS)
without taking into account such unobserved characteristics, the estimated coefficients could be biased.
Therefore, this paper applies the methodological technique of fixed effect panel data which includes
unobserved specific characteristics of municipalities and schools that are fixed across time.
Table 4: Descriptive Statistics of Dependent and Independent Variables of Interest
11 | P a g e
Mean
Standard
Deviation
Minimum
Maximum
Mean
Standard
Deviation
Minimum
Maximum
YEAR
Math_4
Port_4
YEAR
BF_poor
YEAR
Fundef
2005
172.0
165.0
2003
129
2003
426.44
2007
180.5
164.2
2005
1,144
2005
597.33
2009
190.4
172.2
2007
1,550
2006
669.65
2005
15.8
16.9
2003
83
2003
215.36
2007
16.8
16.0
2005
361
2005
291.10
2009
21.3
18.5
2007
340
2006
340.52
2005
115.5
117.4
2003
0.31
2003
0.14
2007
122.3
109.7
2005
23
2005
0.04
2009
134.4
98.7
2007
539
2006
0.02
2005
252.5
238.0
2003
623
2003
5,070.52
2007
290.9
259.0
2005
2,744
2005
6,805.19
2009
314.0
286.7
2007
3,275
2006
9,168.09
YEAR
Math_8
Port_8
YEAR
BF_poor
YEAR
Fundef
2005
232.9
219.2
2003
129
2003
467.67
2007
232.7
222.6
2005
1,163
2005
649.13
2009
234.1
232.4
2007
1,544
2006
727.22
2005
16.5
15.7
2003
84
2003
645.73
2007
16.8
16.3
2005
367
2005
899.00
2009
17.1
17.3
2007
352
2006
1,035.47
2005
180.3
161.6
2003
0.12
2003
0.14
2007
174.7
152.7
2005
23
2005
0.04
2009
172.4
155.1
2007
524
2006
0.02
2005
349.4
310.4
2003
713
2003
41,637.60
2007
347.7
311.4
2005
2,945
2005
56,822.20
2009
344.7
312.7
2007
4,374
2006
65,039.20
This approach can be used to obtain consistent estimators in the presence of omitted variables. The
adopted identification hypothesis is that E(Uit|Xit) = 0, that is, there are no unobserved characteristics fixed
across time from municipalities which are correlated to any independent variable. For instance, important
unobserved factors such as school management may be correlated with Fundef, i.e. spending in elementary
education. Assuming that school management is fixed over time (according to our dataset from 2005 to
2009, this assumption is reasonable to be considered because this is a short period for this type of changes), it
is possible to control for these managerial characteristics, and estimated results will be unbiased. These
estimates are obtained following the functional form expressed by equation (1).
𝑌𝑖𝑡 = 𝛽0 + 𝐵𝐹_𝑃𝑜𝑜𝑟𝑖𝑡−2 𝛽1 + 𝐹𝑢𝑛𝑑𝑒𝑓𝑖𝑡−2 𝛽2 + 𝑆𝑡𝑢𝑑𝑒𝑛𝑡𝑖𝑡 𝛽3 + 𝑆𝑐ℎ𝑜𝑜𝑙𝑖𝑡−2 𝛽4 + 𝑇𝑒𝑎𝑐ℎ𝑒𝑟𝑖𝑡−2 𝛽5 +
𝑀𝑢𝑛𝑖𝑐𝑖𝑝𝑎𝑙𝑖𝑡𝑦𝑖𝑡−2 𝛽6 + 𝐶𝑖 + 𝑈𝑖𝑡
(1)
Where:
12 | P a g e
Yit is the natural log of the students' average score in Prova Brasil (math 4th grade, Portuguese 4th grade, math
8th grade, Portuguese 8th grade);
β0 is the constant of the model;
BF_poorit-2, is the natural log of the ratio, amount of cash allocated through BF divided by the number of
poor people, representing one explanatory variable of interest;
Fundefit-2, is the natural log of the ration, amount spent on elementary education divided by total enrollment
in elementary education, representing another explanatory variable of interest;
Studentit is the vector of control variables with characteristics of the students who took Prova Brasil;
Schoolit is the vector of control variables with characteristics of public schools by municipalities;
Teacherit is the vector of control variables with characteristics of public school teachers by municipalities;
Municipalityit is the vector of control variables with socio-economic characteristics of municipalities;
Ci corresponds to the specific (fixed) effects of each municipality;
Uit is the error term.
The estimations were made using the statistical software Stata 11. Due to the longitudinal feature of
the used data base, the panel data approach was considered in this study. The use of fixed effects for the
regressions was recognized based on both the Breush and Pagan (1980) and Hausman (1978) tests. The
Breush Pagan test indicated that the null hypothesis of non existence of specific effects should be rejected.
Instead, there were indeed unobserved fixed effects in the dataset. Likewise, the Hausman test showed that
the fixed effects were better specifications than the random effects for the dataset used in this study (see
Table 5).
4. ESTIMATION RESULTS
The results for the regressions presented in this section had variables at school and municipal levels,
as displayed in Table 3. The variable Fundef was used to measure public spending on education at the
municipal level. The regression that examined the relationship between quality of education, government
spending on education, and BF program allocation were estimated using two models of estimation: ordinary
least squares for panel data (POLS), random effects, and a fixed effects model. As stated above, the fixed
effect model was the best specification for the dataset examined in this study. Therefore, only the fixed
effects results are presented in Table5. Other estimated coefficients are presented in the appendix.
From Table 5, one can observe a positive significant relationship (p-value < 0.01) between BF_poor
and Math_4and Port_8, showing that this explanatory variable positively affected the average math scores
for4th grade and Portuguese scores for 8th grade. For instance, 1 percent increase in BF allocation per poor
would increase in 0.014 percent the score in math 4th grade and in 0.002 percent the score of Portuguese 8th
grade. Negative significant relationships (p-value < 0.01) existed between BF_poor and Math_8 and Port_4.
For example, 1 percent increase in BF allocation per poor would decrease in 0.003 percent the score in
Portuguese 4th grade and in 0.003 percent the score of math 8th grade. The very small magnitude of the
coefficients may indicate that the increasing enrollment in public schools caused by BF is not affecting
average scores in a relevant manner.
Table 5. Results of natural log of average Prova Brasil scores estimated by fixed-effect panel data
Fixed Effects
Dependent
Variables
13 | P a g e
Math_4
Port_4
Math_8
Port_8
Independent variables of interest
BF_poor
0.012***
0.003***
0.003*** 0.003***
(0.001)
(0.000)
(0.000)
0.016***
-0.001
0.015***
(0.003)
(0.002)
(0.002)
(0.001)
0.037***
(0.003)
Student-related control variables
0.088***
0.050***
Stud_int
(0.008)
(0.008)
-0.164***
Stud_wor
0.192***
(0.007)
(0.007)
0.059***
0.070***
Mother_edu
(0.005)
(0.005)
School-related control variables
0.019***
0.020***
Sch_comp
(0.001)
(0.001)
0.006***
0.009***
Sch_lib
(0.002)
(0.002)
-0.001***
Stu_class
0.001***
(0.000)
(0.000)
Teacher-related control variables
0.030***
0.035***
Teach_edu
(0.003)
(0.003)
Municipality-related control variables
Fundef
GDP
Health_Index
Income_Index
Constant
0.000***
(0.000)
0.109***
(0.009)
0.012**
(0.005)
4.751***
(0.018)
26,499
0.392
0.389
0.39
0.048*** 0.076***
(0.005)
(0.005)
0.069*** 0.147***
(0.004)
(0.005)
0.071*** 0.082***
(0.004)
(0.005)
-0.001
0.014***
(0.001)
(0.001)
-0.001
0.006***
(0.001)
(0.001)
0.000*** 0.000***
(0.000)
(0.000)
-0.003
(0.003)
0.017***
(0.003)
0.000
0.000***
(0.000)
(0.000)
(0.000)
0.085*** 0.013** 0.078***
(0.008)
(0.006)
(0.007)
-0.011**
0.005
-0.001
(0.005)
(0.004)
(0.004)
4.931*** 5.409*** 5.174***
(0.017)
(0.011)
(0.013)
26,499
22,638
22,638
0.152
0.053
0.311
0.448
0.421
0.455
0.366
0.325
0.406
0.000***
Observations
R-square within
R-square between
R-square overall
Breusch-Pagan
2693.90
3062.67 4561.54
χ²(01)
1494.39
2120.67 1524.69
Hausman χ²(11)
Number of
8,833
8,833
7,546
schools
Constants were significant (p<0.01) in all estimations.
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
3077.44
633.82
7,546
Based on these findings, the hypothesis about higher BF allocation coinciding with lower Prova
Brasil average scores is validated just to the early elementary math education (Mathematics 4th grade) and to
the Portuguese 8th grade education. This may be due to the fact that municipalities’ skills to implement BF
vary vastly, indicating that there is not a consistent municipal behavior with regards to BF. As Faria (2002)
14 | P a g e
noted, social programs like the Bolsa Escola(precursor of BF) that require coordination between government
agencies are not being effectively managed by these disparate groups. Differences in competence and
interpretation between agencies can yield inconsistent results in the implementation of these programs. In the
same direction, Rocha (1998) suggested that administrative and coordination problems were “the most
obvious operational shortcoming” in Brazil's social programming. Municipalities face unequal levels of
funding and managerial experience, meaning that they cannot implement these programs equally. There are
abundant examples in the literature of differences in municipal performance at all levels. For example,
Janvry et al. (2010), analyzed how municipal electoral factors affected performance of Bolsa Escola in
reducing school dropout rates of poor children. They observed that municipalities “governed by a first-term
mayor had an estimated 36 percent higher program performance compared to municipalities governed by a
second-term mayor” (p. 3), indicating that political factors can have striking effects on municipal
performance.
When considering Fundef9, there was a positive significant relationship with Math_4, Port_4, and
Port_8(p-value < 0.01), i.e., for these three tests there was a coincidence of higher scores and higher
spending in education. This shows that investments in education are having some positive effects on student
proficiency. For instance, a 1 percent increase in Fundef allocation would increase score in math 4th grade by
0.028 percent and in Portuguese 4th grade and 8th grade by 0.011 percent. Based on these findings, the
hypothesis about higher Prova Brasil scores should be receiving higher government spending on elementary
education was supported. Again, the magnitudes of the coefficients were very small, indicating that the
returns to education that are expected from public spending are not yet very relevant for the Brazilian case.
In general, these results may be related to difference in municipal performances in administering
public spending. There is abundant literature documenting differences in the characteristics and performance
of Brazilian municipalities. Much of this research focuses on the effects of good governance and other
political factors. Motta and Moreira (2009) analyzed how municipal political factors and spending were
related to improvements in social welfare. They found that “the size of the spending needed to attain a
certain level of quality is associated with some characteristics of municipalities, scale, municipal
fragmentation and patronage of an elected governor” (p. 369). In a study of corruption and electoral
accountability, Ferraz and Finan (2009) found that municipal performance varied with electoral incentives
and other related factors. Notably, misuse of funds was more pronounced in municipalities where the mayors
have no reelection incentives, the populace is less privy to information, and where corrupt politicians are less
likely to be punished. Timmons and Garfias (2012) studied the effects of corruption on property tax
collection, a measure of municipal performance. They found a clear relationship between corruption and tax
revenue, where revealed corruption diminished municipal tax revenue and increased the likelihood of
“participatory budgeting”. All of these examples provide evidence for the importance of endogenous factors
in defining how well a municipality provides for its residents.
In Brazil, according to Barros and Mendonça (1997), an improvement in the educational system can
be obtained through the rise of resource allocated to education and/or through the rise of efficiency that they
are used. As spend in education is large in Brazil, the inefficiency can be related to the way that these
resources are allocated. In the present study, we assume that municipal performance in administering public
spending are controlled as a fixed effect. As the obtained results show that the hypothesis about higher BF
allocation coinciding with lower Prova Brasil average scores is validated just to the early elementary math
education (Mathematics 4th grade) and to the Portuguese 8th grade education, municipalities’ skills to
implement BF vary vastly, indicating that there is not a consistent municipal behavior with regards to BF.
9
The dichotomy between poor and rich municipalities led us to explore another variable to measure public spending on education, FINBRA
(Finanças do Brasil, Secretaria Tesouro Nacional) instead of the variable Fundef. Differently from Fundef, FINBRA data are reported by each
municipality annually, including the educational share of their municipal budget. In other words, FINBRA data indicates the intention of
spending in education and they represent a more disaggregated approach (school and municipality levels) with an education public spending
variable that is actually measured. The estimation results for both specifications were consistently very similar, and they are available with the
authors by request.
15 | P a g e
Other secondary results can also be highlighted. When focusing on the student-related control
variables, Stud_int presented a positive significant relationship with all tests (p-value <0.01).These findings
suggest that the higher the economic status of students, the higher their scores. With regards to Stud_wor,
negative significant (p-value <0.01) relationships with all tests were observed. As expected from the literature,
students who work outside the home have a lower proficiency in school. Knowing that one of the reasons BF was
created was to eliminate child labor, a study tracking individual students, instead of schools and
municipalities, would be beneficial to clarify this finding. Mother_edu had a positive significant (p-value <
0.01) relationship with all tests, suggesting that the higher the level of the mother’s education, the higher the
probability of children doing better in school. These finding are in accordance with empirical literature
showing that mothers’ education is one of the most important variables to explain student proficiencies. For
instance, for the 4th grade math regression, an increase of 1% in Mother_edu, corresponded to an increase of
6% in the average test scores.
For school-related control variables, Sch_comp had a positive relationship (p-value < 0.01) with
Math_4, Port_4, and Port_8. Math_8 was negative and not significant. These results are in agreement with
Menezes-Filho’s (2007) findings. He stated that digital inclusion in schools is still a nun clear independent
variable in education literature of education. Menezes-Filho’s (2007) results showed that the presence of
computers had little impact on students’ proficiency and the sign varied: they were occasionally positive, and
sometimes negative. When observing the relationship of Sch_lib, it had a positive (p-value <0.01)
relationship withMath_4, Port_4, and Port_8. These results indicate that school resources, such as the
existence of libraries, were important for improving scores. Considering teacher-related control variables,
Stu_class showed a negative (p-value < 0.01) relationship with all tests. These findings suggest that, in the
Brazilian context, smaller classroom size would help to improve students’ test scores. Teach_edu had a
positive and significant relationship with Math_4, Port_4, and Port_8 (p-value < 0.01). The higher the
educational level of teachers, the better the students’ performance on tests.
With regards to the municipality-related control variables, GDP showed a positive relationship with
Math_4 (p-value < 0.01) and Port_8 (p-value < 0.10). The Health_Index had a positive relationship (p-value
< 0.01) with all tests. To interpret, the healthier the children are in a municipality, the better they do in
school. Income_Index had a positive significant (p-value <0.01) relationship only with Math_4.This means
that students in municipalities with a higher the economic status, did better on the tests.
4.
CONCLUSION
The main objective of this paper is to examine whether Brazil’s scarce public resources were being
allocated in a coordinated manner during the period 2003-2009. The overall research question it attempts to
answer is as follows: what are the relationships between quality of education, educational spending, and BF
cash transfer? Prova Brasil, a national test administered by the Brazilian Ministry of Education, is used as a
proxy for measuring quality of education. Higher average scores in Prova Brasil indicate better quality of
education. Fundef data is used to measure public spending on education. The research question was answered
using regressions that treated Prova Brasil scores as the dependent variable and BF and Fundef spending
(along with numerous controls) as the independent variables. The results of the regression also allowed us to
address the paper’s two research hypotheses: 1) municipalities with higher level of BF allocation also had
lower scores in Prova Brasil; and 2) municipalities characterized by higher Prova Brasil scores also received
higher government spending on elementary education. The first hypothesis is based on the fact that BF is
increasing enrollment in public schools, but the public school system may not be able to provide resources
needed to meet the needs of more students. The second hypothesis is based on the idea that higher
government spending on education should result in better education performance.
16 | P a g e
Results indicated that the hypothesis about higher BF allocation coinciding with lower Prova Brasil
average scores is validated just to the early elementary math education (Mathematics 4th grade) and to the
Portuguese 8th grade education. The very small magnitude of the coefficients may indicate that the increasing
enrollment in public schools caused by BF is not affecting average scores in a relevant manner. For instance,
1 percent increase in BF allocation per poor would increase in 0.014 percent the score in math 4 th grade and
in 0.002 percent the score of Portuguese 8th grade. Negative significant relationships (p-value < 0.01) existed
between BF_poor and Math_8 and Port_4. For example, 1 percent increase in BF allocation per poor would
decrease in 0.003 percent the score in Portuguese 4th grade and in 0.003 percent the score of math 8th grade.
This may be due to the fact that municipalities’ skills to implement BF vary vastly, indicating that there is not
a consistent municipal behavior with regards to BF.
When considering Fundef, the hypothesis about higher Prova Brasil scores should be receiving higher
government spending on elementary education was supported. The investments in education are having some
positive effects on student proficiency. For instance, a 1 percent increase in Fundef allocation would increase
score in math 4th grade by 0.028 percent and in Portuguese 4th grade and 8th grade by 0.011 percent. Based
on these findings, the hypothesis about higher Prova Brasil scores should be receiving higher government
spending on elementary education was supported. Again, the magnitudes of the coefficients were very small,
indicating that the returns to education that are expected from public spending are not yet very relevant for
the Brazilian case. Again, the magnitudes of the coefficients were very small, indicating that the returns to
education that are expected from public spending are not yet very relevant for the Brazilian case.
In general, these results may be related to difference in municipal performances in administering
public spending. Some of the potential limitations of this study follow. First, fixed effects approach may not
account for endogeneity between BF allocation and/or public spending in education and unobserved
variables. Second, understanding the positive relationship between test scores and child labor is difficult at
the level of analysis utilized here. Future research should focus on using individual students as the unit of
analysis instead of schools and shed some light on this issue. Unfortunately, current limitations on data
availability precluded such an analysis.
REFERENCES
Acemoglu, D. (1996). A microfoundation for social increasing returns in human capital accumulation. The
Quarterly Journal of Economics, 120, 779-804.
Acemoglu, D., 1998. Why do new technologies complement skills? Directed technical change and wage
inequality. Quarterly Journal of Economics, 113, 1055–1090.
Aghion, P., & Hewitt, P. (1992). A Model of Growth through Creative Destruction. Economettica, 60(2). 323–
351.
Baldacci, E., Clements, B., Gupta, S., & Cui, Q. (2008). Social Spending, Human Capital, and Growth in
Developing Countries. World Development, 36(8), 1317-1341.
Barro R. J. (2001). Human capital and growth. American Economic Review, 91(2), 12- 17.
Barros, R. P. de, Mendonça, R.S.P.;O impacto de gestão escolar sobre o desempenho educacional.
Washington: BID, 39 p, 1997.
Benhabib, J., & Spiegel M. M. (2005). Human capital and technology diffusion. Handbook of Economic Growth,
Amsterdam: North Holland, 935-966.
Blankenau, W. F., Simpson, N. B. (2004). Public education expenditures and growth. Journal of Development
Economics, 73, 583-605.
Bourguignon, F., Ferreira F. H. G., & Leite P. G. (2003). Conditional cash transfers, schooling, and child labor:
micro-simulating Brazil’s bolsa escola program. The World Bank Economic Review, 17(2), 229–254.
Cardoso, E. & Souza, A. P. (2004). The Impact of Cash Transfers on ChildLabor And School Attendance In
Brazil, Vanderbilt University, Nashville Tennessee.
17 | P a g e
Ciccone, A., & Peri, G. Indentifying human-capital externalities: Theory with applications. The Review of
Economics Studies, 73(2), 381-412.
Davies, N. (2006). FUNDEB: A Redenção da Educação Básica? Educ. Soc., Campinas, vol. 27, n. 96 Especial, p. 753-774, out. 2006.
Faria V (2002) Institutional reform and government coordination in Brazil’s social protection policy. ECLAC
Review 77: 7–24.
Ferro, A.R.; & Kassouf, A.L. (2005). Avaliação do impacto do Programa Bolsa-Escola sobre o trabalho infantil
no Brasil. Pesquisa e Planejamento Econômico, 35(3), 417-444.
Fleisher, B. M., Hu, Y., Li, H., & Kim, S. (2011) Economic transition, higher education and worker productivity
in China. Journal of Developmente Economics. 94, 86-94.
Glewwe, P.; & Kassouf, A. L. (2008). O impacto do Programa Bolsa Família no total de matrículas do ensino
fundamental, taxas de abandono e aprovação. In: Anais do Encontro Nacional de Economia, 36. Salvador.
Gordon, Nora and Vegas, Emiliana. “Educational Finance Equalization, Spending, Teacher Quality, and
Student Outcomes: The Case of Brazil’s FUNDEF,” in Incentives to Improve Teaching: Lessons from Latin
America, edited by E. Vegas (Washington: World Bank Press, 2005).
Haddad, M. A. (2008). Bolsa Família and the needy: is allocation contributing to equity in Brazil? Journal of
International Development, 20, 654-669
Hanushek, E. A., & Wossmann. (2007). The Role of Education Quality in Economic Growth. World Bank Policy
Research Working Paper, No. 4122. Washington, DC: World Bank.
Janvry, Alain, Finan, F. e Sadoulet, Elisabeth. (2006), “Evaluating Brazil’s Bolsa Escola Program: Impact on
Schooling and Municipal Roles”. Manuscrito.
Landim Junior P.H. (2009). Os Efeitos do Programa Bolsa Familia sobre a Economia dos Municipios Brasileiros.
INSPER.
Lin, T. C. (2003). Education, technical progress, and economic growth: the case of Taiwan. Economics of
Education Review. 22, 213-220.
Liso, H. A. (2010). O Impacto do Programa Bolsa Família nos Indicadores de Desempenho Escolar. INSPER.
Lucas, R. E. (1988). On the mechanics of economic development. Journal of Monetary Economics, 22, 3-42.
Mankiw, N. G., Romer D., & Weil D. (1992). A contribution to the empirics of economic growth. Quarterly
Journal of Economics 107(2), 407-437.
Mello, L., & Hoppe, M. (2005). Education attainment in Brazil: The experience of FUNDEF. Economics
Department Working Papers No.424. OECD, Paris.
Menezes-Filho, N. (2007). Os determinantes do desempenho escolar do Brasil. São Paulo: IBMEC. Sumário
Executivo.
Menezes-Filho, N. and Pazello, E. (2007).Do teachers’ wages matter for proficiency? Evidence from a
funding reform in Brazil. Economics of Education Review. Volume 26, Issue 6, December 2007, Pages 660–
672.
Motta, Ronaldo Seroa da; Moreira, Ajax. Political factors and the efficiency of municipal expenditure in
Brazil. Econ. Apl., Ribeirão Preto, v. 13, n. 4, Dec. 2009 .
Nelson R. R., Phelps E. (1966). Investment in humans, technology diffusion and economic growth. American
Economic Review 56(2), 69-75.
Oketch, M. O. (2006). Determinants of human capital formation and economic growth of African countries.
Economics of Education Review. 25, 554-564.
Park, T-Y., Choung, J-Y., & Min, H-G. The cross-industry spillover of technological capability: Korea`s DRAM
and TFT-LCD industries. World Development, 36(12), 2855-2873
Pedrozo, E. (2007). Uma avaliação ex-ante dos impactos do Bolsa Família na redução do trabalho infantil. In:
Anais do XXXII Encontro Nacional de Economia.
Ranis, G., Stewart, F., & Ramirez, A. (2000). Economic growth and human development. World Development,
28(2), 197-2000.
18 | P a g e
Rocha, S. Pobreza no Brasil ⎯ principais tendências da espacialização. XXXVI Encontro Nacional de
Economia, Vitória, dez. 1998, Anais, p. 1665-1682.
Romero, J. A. R, & Hermeto A. M. (2009). Avaliação de Impacto do Programa Bolsa Família sobre Indicadores
Educacionais: Uma Abordagem de Regressão Descontínua. Cedepar.
Soares, F. V., Soares, S., Medeiros, M., & Osório, R. G. (2006) Cash transfer programs in Brazil: Impacts on
inequality and poverty. International Poverty Center Working Paper No. 21. Brasilia, DF: United Nation
Development Program.
Tavares, P.A.; Pazello, E.T.; Camelo, R.S.; & Fernandes, R. (2009). Uma avaliação do Programa Bolsa Família:
focalização e impacto na distribuição de renda e pobreza. Pesquisa e Planejamento Econômico, 39.
Teixeira, A., & Fortuna, N. (2004). Human capital, innovation capability and economic growth in Portugal, 19602001. Portuguese Economic Journal, 3(3), 205–225.
Teles, V. K., & Andrade J. (2008). Public investment in basic education and economic growth. Journal of
Economic Studies, 35(4),. 352-364.
Vale, T. M. C., Felipe, J. L. A., Formiga, M. C. C., & Ramos P. C. F. ( 2010). Redução da pobreza e
transferências governamentais: um estudo de caso para o estado Rio Grande do Norte na região Nordeste
brasileira. XVII Encontro Nacional de Estudos Populacionais, ABEP.
Wolff, E. N. Human capital investment and economic growth: Exploring the cross-country evidence. Structural
Change and Economic Dynamics. 11, 433-472.
APPENDIX
Table A.1: Descriptive statistics of independent variables (cont.)
4th grade
Variable
8th grade
BF_poor
6.30
Standard
deviation
1.32
Fundef_d
6.23
0.49
-4.19
9.12
6.28
0.52
-4.19
11.08
Stud_int
0.13
0.13
0.00
0.86
0.17
0.16
0.00
0.91
Mean
Min
Max
6.31
Standard
deviation
1.32
-2.10
8.38
Min
Max
Mean
-1.16
8.09
Stud_wor
0.15
0.08
0.00
0.76
0.19
0.10
0.00
1.00
Mother_edu
0.81
0.12
0.00
1.00
0.82
0.12
0.00
1.00
Sch_comp
0.33
0.47
0.00
1.00
0.60
0.49
0.00
1.00
Sch_lib
0.34
0.47
0.00
1.00
0.45
0.50
0.00
1.00
Stu_class
30.45
6.80
1.00
114.00
34.04
7.36
1.00
112.33
Teach_edu
0.59
0.34
0.00
1.00
0.79
0.27
0.00
1.00
GDP
10428.13
11964.46
1124.07
389828.80
11905.57
12321.93
1124.07
389828.80
Health_Index
0.69
0.12
0.03
0.96
0.71
0.13
0.03
0.99
Income_Index
0.51
0.22
0.00
0.99
0.54
0.23
0.00
0.99
Note: The number of observations to the pooled data (2005, 2007 and
2009) to the 4th grade is 26,499 and to the 8th grade is 22,638.
19 | P a g e
Table A.2. Results of natural log of average Prova Brasil scores estimated by Polled OLS
Dependent Variables
lnBF_poor
lnFundef_d
Stud_int
Stud_wor
Mother_edu
Sch_comp
Sch_lib
aluno_turma_4
Teach_edu
GDP
Health_Index
Income_Index
LAT
LON
(1)
Math_4
(2)
Port_4
(3)
Math_8
(4)
Port_8
0.018***
(0.000)
-0.001
(0.001)
0.242***
(0.006)
-0.300***
(0.008)
0.093***
(0.005)
0.001
(0.001)
0.008***
(0.001)
-0.001***
(0.000)
0.013***
(0.002)
-0.000***
(0.000)
0.074***
(0.007)
-0.021***
(0.003)
-0.003***
(0.000)
-0.001***
(0.000)
-0.000
(0.000)
-0.005***
(0.001)
0.245***
(0.006)
-0.348***
(0.008)
0.129***
(0.005)
-0.001
(0.001)
0.012***
(0.001)
-0.001***
(0.000)
0.008***
(0.002)
-0.000***
(0.000)
0.063***
(0.007)
-0.015***
(0.003)
-0.003***
(0.000)
-0.001***
(0.000)
-0.003***
(0.000)
0.002***
(0.001)
0.179***
(0.004)
-0.049***
(0.005)
0.104***
(0.004)
-0.006***
(0.001)
0.012***
(0.001)
0.005***
(0.000)
0.003***
(0.001)
0.171***
(0.004)
-0.121***
(0.005)
0.110***
(0.005)
0.004***
(0.001)
0.011***
(0.001)
4.908***
(0.010)
4.960***
(0.010)
0.014***
(0.002)
-0.000***
(0.000)
0.040***
(0.005)
-0.035***
(0.003)
-0.001***
(0.000)
-0.001***
(0.000)
-0.001***
(0.000)
5.293***
(0.007)
0.016***
(0.002)
-0.000***
(0.000)
0.065***
(0.005)
-0.019***
(0.003)
-0.000*
(0.000)
-0.001***
(0.000)
-0.001***
(0.000)
5.167***
(0.007)
26,499
0.482
22,638
0.410
22,638
0.457
aluno_turma_8
Constant
Observations
26,499
R-squared
0.502
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
20 | P a g e
Table A.3. Results of natural log of averageProva Brasil scores estimated by random effects panel data
Variables
lnBF_poor
lnFundef_d
Stud_int
Stud_wor
Mother_edu
Sch_comp
Sch_lib
aluno_turma_4
Teach_edu
GDP
Health_Index
Income_Index
(1)
Math_4
(2)
Port_4
(3)
Math_8
(4)
Math_8
(5)
Port_8
0.014***
(0.000)
0.008***
(0.001)
0.256***
(0.006)
-0.260***
(0.006)
0.087***
(0.004)
0.010***
(0.001)
0.003***
(0.001)
-0.001***
(0.000)
0.017***
(0.002)
0.000
(0.000)
0.125***
(0.006)
-0.006*
(0.003)
-0.004***
(0.000)
0.001
(0.001)
0.242***
(0.006)
-0.296***
(0.006)
0.114***
(0.004)
0.010***
(0.001)
0.006***
(0.001)
-0.001***
(0.000)
0.016***
(0.002)
0.000
(0.000)
0.115***
(0.006)
-0.006*
(0.003)
-0.005***
(0.000)
0.004***
(0.001)
0.149***
(0.003)
-0.053***
(0.004)
0.103***
(0.004)
-0.002**
(0.001)
0.000
(0.001)
-0.005***
(0.000)
0.004***
(0.001)
0.149***
(0.003)
-0.053***
(0.004)
0.103***
(0.004)
-0.002**
(0.001)
0.000
(0.001)
0.004***
(0.000)
0.006***
(0.001)
0.147***
(0.004)
-0.133***
(0.004)
0.116***
(0.004)
0.009***
(0.001)
0.008***
(0.001)
4.920***
(0.010)
4.988***
(0.010)
0.009***
(0.002)
-0.000*
(0.000)
0.055***
(0.005)
-0.009***
(0.003)
-0.000***
(0.000)
5.332***
(0.007)
0.009***
(0.002)
-0.000*
(0.000)
0.055***
(0.005)
-0.009***
(0.003)
-0.000***
(0.000)
5.332***
(0.007)
0.016***
(0.002)
-0.000*
(0.000)
0.066***
(0.005)
-0.010***
(0.003)
-0.000***
(0.000)
5.210***
(0.007)
26,499
26,499
22,638
22,638
22,638
8,833
8,833
7,546
7,546
7,546
aluno_turma_8
Constant
Observations
Number of
ESCOLA_CODIGO
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
21 | P a g e